pre-commit with ruff

  pre-commit with ruff

  - Add 22 unit tests for parse_args.py covering VariableRecovery, LayersContext,
  and parse_args
  - Add type annotations to parse_args.py (return types, proper Optional types)
  - Configure ruff following cvxpy's pragmatic approach (E, F, I rules only, 100
  char line length)
  - Enable PT (pytest-style) rules for tests with proper match parameters
  - Fix all PT011/PT012 violations in existing tests
  - Add .pre-commit-config.yaml with ruff linter/formatter
  - Exclude mpax_if.py temporarily (TODO: fix separately)

  - Add 22 unit tests for parse_args.py with full coverage of edge cases
  - Add proper type annotations to parse_args.py
  - Configure ruff following cvxpy's pragmatic approach (100 char lines, E/F/I
  rules)
  - Enable PT (pytest-style) rules and fix all 13 violations in existing tests
  - Enable pyright type checker (scoped to new files only for incremental
  adoption)
  - Fix type issues: solver parameter accepts None, q can be None
  - Exclude mpax_if.py from ruff temporarily
  - Add pre-commit hooks: ruff, pyright, standard file checks

* fixing tests

* working on batching

* batching works

* fixed tests

* get CI passing

* install cvxpy from mater in CI

* up CI to ptyhon 3.12

Adds jax interface mirroring torch interface

Add GP support

…el frameworks

…ers-new

- Remove ~160 lines of dead code:
  - _MxTrackedArray class (unused numpy subclass)
  - _mx_numpy_array, _mx_numpy_eye, _mx_numpy_solve (never called)
  - _patch_mx_grad and _reduce_to_scalar (global mx.grad monkey-patching)
  - Unused _original_* variables and restore statements
- Align structure with torch/jax implementations:
  - Same helper functions with consistent docstrings
  - Rename _scipy_csr_to_numpy_array -> _scipy_csr_to_dense
  - Rename _solve_canonical -> _solve_with_vjp
  - Use mx.broadcast_to instead of custom _expand_unbatched_param
- Fix tests to use mx.grad correctly:
  - Add argnums=(0,1) for multi-arg gradients
  - Ensure scalar return values

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

Apple MLX tensors support (refactor according to code of other frameworks of cvxpylayers1.0 )

* loosening dependencies

* bump to python 3.10

* minimize delta

* Adds shape tests

* Rewrites tests

* adds Jax 0.6.0 support

* fixed uv sycn

* minor fixes

* Clean rewrite of Moreau interface with zero-copy GPU tensors

- Use moreau_torch for PyTorch path (zero-copy GPU operations)
- Remove diffqcp/cupy dependencies - gradients raise NotImplementedError
- Create solver once in MOREAU_ctx with dynamic batch size
- torch_to_data handles GPU tensor prep using torch indexing
- Split into MOREAU_data (torch) and MOREAU_data_jax classes
- Add test for backward NotImplementedError

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* caching solver in cvxpylayers

* update moreau interface

* Add CPU support to moreau interface

- Replace hard-coded .cuda() calls with device-aware .to(device, dtype)
- Add get_cpu_solver() for numpy-based moreau.Solver on CPU tensors
- CUDA path: zero-copy GPU operations with moreau.TorchSolver
- CPU path: zero-copy torch<->numpy conversion with moreau.Solver
- Add CPU-specific tests (equality, mixed constraints, batched)
- Add device consistency tests (output matches input device)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* align moreau interface with new API

* cleaned up depedencies

* update moreau API

* support settings

* Alternate CuClarabel+diffqcp implementation (#182)

* initial work; not finished--saving progress

* more work; almost finished with cuclarabel first pass

* done? saving progress. now to test

* cleaning up

* bugfixes

* working status

* Various fixes for diffqcp

* quad obj is None handling

* bugfixes + consolidate cuclarabel

* remove old cuclarabel

* cleaning up before opening PR

* more cleaning

* re-add lock file

* addressing PR comments

* make cuclarabel test use double precision

* let there be doubles

---------

Co-authored-by: Parth Nobel <parthnobel@berkeley.edu>

* Formats cuclarabel_conif.py

* Adds shape tests (#185)

* Adds shape tests

* Rewrites tests

* Adds torch example

* Adds jax_example

* Updates README.md

* SPEC0 alignment in Python versioning

---------

Co-authored-by: Steven Diamond <diamond.po.central@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>

* first attempt at fixes

* depedency stuff

* small fix + readme

* switch to lsmr default diffqcp